Qubit Calibration Pipelines
A calibration pipeline is the ordered set of automated experiments that turn a cold, uncharacterized chip into a device with accurate gates.
Why Calibration Is a Pipeline
A superconducting processor arrives at base temperature with only approximate knowledge of each qubit's frequency, each pulse's correct amplitude, and each readout's threshold. Dozens of parameters per qubit must be measured, and they depend on one another: you cannot calibrate a two-qubit gate before single-qubit gates work, and you cannot tune single-qubit gates before you know the qubit frequency and can read it out. Calibration is therefore a directed sequence, a pipeline, in which each step depends on earlier ones.
A Typical Order
- Find each resonator and its readout frequency by spectroscopy.
- Find each qubit frequency, then optimize the readout point and thresholds.
- Calibrate pi and pi-over-two pulse amplitudes and the DRAG coefficient.
- Fine-tune drive frequency with Ramsey experiments; measure coherence times.
- Calibrate two-qubit gates, then cancellation tones and phase corrections.
Dependency Management
Because parameters drift and depend on one another, modern calibration is organized as a graph of experiments with dependencies, sometimes called an optimization graph. When a downstream check fails, the system walks back up the graph to recalibrate the parameters it depends on, rather than blindly rerunning everything. This keeps a large device calibrated with the least measurement time.
Coarse and Fine Steps
Many parameters use a two-stage strategy: a coarse measurement to get within range, then an error-amplifying fine measurement that repeats an operation many times so a small error accumulates into a large, easily measured signal. Amplitude calibration by repeated pi pulses is the classic example. This structure gives both robustness and precision.
Drift and Automation
Qubit parameters drift on timescales from minutes to days, driven by temperature, flux, and defect dynamics. Practical systems recalibrate continuously or on a schedule and monitor benchmarking metrics to decide when recalibration is needed. As device size grows, fully automated, dependency-aware calibration becomes essential, because manual tuning of thousands of parameters is not feasible.
The output of the pipeline is validated by randomized benchmarking and related techniques described elsewhere.